Multiobjective Water Quality Model Calibration Using a Hybrid Genetic Algorithm and Neural Network-Based Approach

This study aims at coupling a hybrid genetic algorithm (HGA) and a neural network (NN) model for the multiobjective calibration of surface water quality models. The HGA is formed as a robust optimization algorithm through combining a real-coded genetic algorithm with a local search method, i.e., the Nelder-Mead simplex method (NMS). The NN model is developed to approximate the input-output response relationship underlying a numerical water quality model, and is then incorporated into the HGA framework, which results in the HGA-NN approach. This approach has the advantage of evaluating the objective function of the calibration model in a more efficient way. The HGA-NN approach is tested in the calibration of a CE-QUAL-W2 model which is set up to simulate the hydrodynamic process and water quality conditions in Lake Maumelle in central Arkansas. It is found that the HGA-NN approach can improve the computational efficiency. However, it does not guarantee the finding of the parameter values with a low objective function value. An adaptive HGA-NN approach is then proposed to improve its performance. In this adaptive approach, both the water quality model and the NN model are incorporated into the HGA framework. They are executed adaptively to evaluate the objective function. The application results demonstrate that the adaptive approach can be applied to the calibration of water quality models.